XDOF (pronounced "ecks-doff") builds the data supply chain for robotics AI — the layer that frontier AI labs and humanoid robot companies need but won't build themselves. Founded in October 2024 by UC Berkeley robotics researchers, the company runs a three-tier data collection architecture: deployment-robot teleoperation in warehouse-scale environments, GELLO-device teleoperation (a low-cost system the founders invented), and egocentric wearable sensors that capture everyday human motion. Raw data feeds into annotation pipelines, cleaning tools, and an exabyte-scale data warehouse, creating a continuous quality loop. The output is high-fidelity manipulation and locomotion training data for foundation models. XDOF also released ABC-130K — claimed as the world's largest open-source bimanual robot manipulation dataset, with 130,000 trajectories across 195 tasks — built in partnership with UC Berkeley, CMU, MIT, and Amazon. Customers are frontier AI labs and humanoid robot companies; the company had 20 paying customers and ~60 employees at the time of its June 2026 public launch. The name is a riff on "degrees of freedom" in robotics — X meaning any or infinite.